Speeding-related crashes continue to be a serious problem in the United States. The proportion of speeding-related fatal crashes has changed little in over a decade. In 2012, 30% of all fatal crashes had speeding as a contributing factor (NCSA, 2014), the same percentage as in 1996 (Liu, Chen, Subramanian, & Utter, 2005). Speeding is a complicated behavior that varies by driver and situation (Richard et al., 2013a). It is also a common behavior, with most drivers reporting that they drive over the speed limit at least some of the time (Schroeder, Kostyniuk, & Mack, 2012). Given the widespread occurrence of speeding and the high toll in injuries and lives lost in speed-related crashes, 10,219 fatalities in 2012 (NCSA, 2014), as well as the high economic costs of speed-related crashes (Blincoe et al., 2014), this is a safety issue that demands a great deal of attention. The recently completed NHTSA project, Motivations for Speeding (Richard et al., 2013a), collected data that can be used to address questions about driver speeding behavior. This was a naturalistic driving study in which 1-Hz Global Positioning System (GPS) data were collected from 88 drivers in Seattle and 76 drivers in rural Texas (College Station) to record how fast vehicles traveled on different roadways. This effort resulted in a rich and unique data set that can provide a wide range of insights into speeding behavior. The current project further developed this data set and conducted additional analyses to expand upon the information already generated from this data set. A key activity was to operationalize speeding in terms of individual Speeding Episodes that capture driving characteristics during a particular speeding event. The specific study objectives in this research were to: 1. Redefine speeding in terms of Speeding Episodes and use the new data to identify underlying types of speeding and Driver Types. 2. Conduct additional data analyses on the relationships between situational factors and speeding. A key goal of the current research was to redefine speeding behavior in terms of holistic speeding events. To support this definition of speeding, time series of GPS driving data were parsed into Free-flow Episodes (FFEs), and Speeding Episodes (SEs), which represent short yet continuous segments or “snippets” of driving time extracted from a trip. Free-flow Episodes are used as a proxy for a driver’s opportunity to speed within a trip, and they exclude parts of a trip where a vehicle is stopped, trapped in traffic congestion, slowing to a stop for a traffic control device, etc.–situations in which it would be unlikely that a driver has a chance to speed. Speeding Episodes are used as the primary data element in the analysis of speeding, and they represent an interval of driving in which a vehicle is potentially speeding. In the current study, speeding conceptually represents a risk of getting a speeding ticket more than a clear risk of getting into a speed-related crash. The operational definitions of these data elements were as follows: • A Free-Flow Episode (FFE), which represents the opportunity to speed, occurred when the driver was traveling at or above a threshold set at 5 mph below the posted speed limit. This speed criterion had to be maintained for at least 30 seconds to be included as a FFE. • A Speeding Episode (SE) was defined as continuous driving at or above a threshold set at 10 mph above the posted speed limit. This speed criterion had to be maintained for at least 6 seconds to be included as a SE. Time-series data were reduced into data structures summarizing trip, FFE, and SE characteristics. Examples of variables included in this reduction include descriptive statistics for speed and acceleration, duration, and percent time on a particular road type. Geographic Information System (GIS) maps of the Seattle and Texas sites were used to obtain posted speed limit, functional class, and other available road network data, which then were linked to driving data. In Seattle, there were a total of 4,754 separate SEs recorded across 6,137 trips with at least one FFE. In Texas, there were a total of 1,376 separate SEs recorded across 4,645 trips with at least one FFE. The underlying premise in these analyses is that speeding results from qualitatively different types of driving behaviors, or from different speeding choices. In this case, it is possible that SEs may take different forms depending on the underlying motivations or associated behaviors. Thus, a key objective of this project was to identify different types of speeding. Cluster analyses were conducted to answer specific questions about driver speeding behavior. All analyses described below were conducted on variables calculated using time-series data within individual SEs (e.g., mean speed, total duration, etc.). In addition, cluster analyses were conducted separately for Seattle and rural Texas data because the roadway environments and driving patterns at these locations differed substantially. The specific speeding-behavior questions addressed with the cluster analysis included: • Is it possible to characterize types of speeding based on characteristics of Speeding Episodes? • Is it possible to classify subtypes of speeders (Driver Types) using patterns in their types of speeding? • To what extent are Driver Types defined by demographics and/or attitudes and beliefs about speeding? The basic process used to address these questions involved three steps. The first step was to conduct a cluster analysis on select SE variables to divide SEs into groups/types of speeding based on shared characteristics. The second step involved identifying relatively how often each driver engaged in the different types of speeding, thus creating a “speeding profile” for each driver. These speeding profiles were then used as the basis for a second cluster analysis to identify different Driver Types based on similarities in speeding profiles. The final step involved running a Chi-square test on the Driver Types to determine if category membership reflects demographics. In addition to this, patterns in responses to personal inventory items about attitudes and beliefs about speeding were examined for trends across Driver Types. Seattle Cluster Analyses for Speeding Type Variables representing different characteristics of SEs (e.g., duration, mean speed, etc.) were entered into a k-means cluster analysis. The most interpretable result was the 6-cluster solution described below: Cluster 1 - Speeding Up: The characteristics of these SEs are consistent with speeding up that can occur immediately prior to an increase in the posted speed. Cluster 2 — Speed Drop: The characteristics of these SEs are consistent with drivers slowing down after transitioning to a lower posted-speed zone. Cluster 3 — Incidental Speeding: This is the most common type of speeding, and it likely represents unintentional or incidental speeding on the part of a driver. Cluster 4 — Casual Speeding: This type of speeding is similar to Incidental speeding, but differs primarily in degree, with all variables having higher values. It seems to represent a more accepting or casual attitude towards speeding in the situations in which they occur. Cluster 5 — Cruising Speeding: The defining aspect of this type of speeding is the long duration relative to the other types, akin to drivers “cruising” along a roadway at elevated speeds for a moderate duration. Cluster 6 — Aggressive Speeding: This type of speeding represents more aggressive and/or riskier driving than other types. Overall, the cluster analysis was effective at parsing SEs into groups whose members were similar across multiple dimensions. More importantly, the resulting clusters are interpretable. Seattle Cluster Analysis to Identify Driver Types A second cluster analysis was conducted to determine if it is possible to classify individuals into different Driver Types with respect to speeding. Table ES-1 provides the median values for the proportion of each type of speeding by Driver Type. The following sections summarize the characteristics of the Driver Types described above and provide the rationale for the labels assigned to each Driver Type. Driver Type 1 — Deliberate: Drivers in this group had relatively more Casual, Cruising, and Aggressive SEs, but fewer Incidental SEs. Moreover, the Cruising and Aggressive types of speeding are the ones that most likely represent deliberate attempts to speed, and at relatively high speeds as well. Individuals in this group also had substantially more SEs than those in other groups. This group seemed to represent drivers that are most willing to engage in deliberate or intentional speeding behavior (i.e., with longer durations and at higher speeds). Driver Type 2 — Typical Speeders: Drivers in this group are labeled as the Typical Driver Type primarily because, as Table 8 shows, the distribution of SEs closely matched the average distribution across all Driver Types. This group contained the largest number of drivers. Individuals in this Driver Type also occupied a middle range in terms of their speeding profiles and the overall frequency of SEs. Driver Type 3 — Situational Speeders: This group is overrepresented in terms of the Speeding Up type of speeding (16% vs 5% on average) and, like the drivers in the Deliberate Driver Type, they have proportionately less Incidental speeding. The Speeding Up type occurred in specific situations, such as in advance of speed limit increases, and in sections where there seemed to be a mismatch between the roadway cues and the posted speed limit. Driver Type 4 — Unintentional Speeders: In contrast to the other Driver Types, this last group was made up of drivers that mostly engaged in Incidental speeding, combined with a small amount of Casual speeding. The other types of speeding were uncommon in this group. Extent to which Driver Types were Defined by Demographics Another analysis was conducted on Seattle SE data to determine the extent to which Driver Types were defined by demographics. The previous cluster analyses made it possible to assign a Driver Type (using cluster membership) to individual drivers based upon the types of speeding events that they incurred (see Table ES-2). A chi-squared test was run on the results to determine if any significant differences in group membership could be found between demographic groups. The chi-squared test with 9 degrees of freedom had a p-value of less than 0.001, which indicated that a significant difference was found in cluster memberships across demographic groups. Situational and Unintentional Driver Types had the highest frequency of Older drivers (male and female), while Typical Driver types were more frequently Younger females and Deliberate Driver Types were more frequently Younger Males; however, this difference was a matter of degree. All demographic groups were found in each driver type. Trends in Beliefs and Attitudes across Driver Types Participants in the original Motivations for Speeding study completed multiple personal inventory questions about their attitudes, motivations, and beliefs towards speeding (Richard et al., 2013b). Responses to these questions provided a way to obtain further insight about how the Driver Type groups may differ. Mean values for each question were calculated for all members of a particular Driver Type. The objective was to identify patterns in responses that might provide qualitative insight about how the Driver Types differed in terms of their attitudes and beliefs about speeding. Table ES-3 below summarizes the trends observed across questions. A few consistent patterns occurred across the different types of questions in the surveys. The dominant trend was that responses from the Deliberate speeding driver group were on the aggressive end of the spectrum relative to other Driver Types. This pattern was evident for almost all questions. Another less prominent trend that occurred within certain sets of questions was that Unintentional Driver Types were on the more conservative/safer end of the response spectrum. Neither of these trends were surprising given the types of speeding that defined these Driver Types. With regard to the Typical and Situational Driver Types, they were usually not different from the Unintentional Driver Type, and they sometimes fell in between the Deliberate and Unintentional Driver Types in their responses to these questions. For almost all questions, differences between Typical and Situational Driver Types were minimal. Texas Cluster Analyses for Type of Speeding The data from Texas were analyzed in a manner parallel to that of the Seattle data. Most of the clusters identified for the Texas data were comparable to particular types of speeding clusters in Seattle. The only exception was Cluster 6. Rather than representing an Aggressive type of speeding, this cluster (Small Increase) seems to represent a shortened version of the Speeding Up cluster. Note that the Driver Type cluster analysis was not run in Texas. This is because over a quarter of Texas drivers had an insufficient number of SEs to compute the proportional distribution of types of speeding for those drivers. Situational Analysis To obtain a better understanding of situational aspects of speeding, we used road-network data to examine where different types of Speeding Episodes occurred in both Seattle and Texas. Specifically, SEs were mapped to identify the locations where SEs occurred more frequently to determine if there were roadway characteristics that systematically encouraged different types of speeding (identified as speeding “hot spots”). To do so, we employed an exploratory approach of using the proportion of SEs relative to FFEs at various locations to develop speeding “heat maps.” The goal of this analysis was to determine if there were patterns with respect to where SEs occurred that could provide insight into the different speeding types identified in the cluster analysis. The findings regarding the different types of speeding are summarized below. Speeding Up: These SEs mostly occurred on arterials and were tied to transition zones or roads with speed limit changes that are incongruent with perceived design speed. Speed Drop: These SEs mostly occurred on arterials and were tied to speed limit transition zones. Incidental: These SEs occurred across widespread regions of the road network. There were clear “hot spots,” which may indicate locations in which the driving environment encouraged speeding, even when drivers were not intending to speed. Casual: These SEs were slightly less widespread than Incidental speeding, but with fewer “hot spots.” Cruising: The majority of the SEs in the Cruising cluster occurred on the freeways and state highways. Aggressive (Seattle Only): These SEs were distributed throughout the map with only a few “hot spots.” This type of speeding also had the highest prevalence of riskier speeding characteristics, such as exceeding the speed limit by 20 mph and more speeding at night. The link between roadway characteristics and SEs seemed to be the weakest for Aggressive SEs, which suggested that driver-specific factors may play a more important role in this type of speeding. Small Increase (Texas Only): These SEs were mostly tied to speed limit transition zones on lower-speed roads. There were two primary objectives in this project. The first was to redefine speeding in terms of Speeding Episodes and use the new data to identify underlying types of speeding and Driver Types. The second objective was to conduct additional data analyses on the relationships between situational factors and speeding. Overall, we were successful in accomplishing the first objective in full; however, limited data on situational factors permitted us to only address the second objective at a high level. The key conclusions are described below. The Cluster Analysis Approach was Useful for Identifying Types of Speeding A primary objective in this project was to advance beyond the current notions of speeding as a monolithic and aggregate concept and develop a more nuanced understanding of the behavioral factors that comprise speeding. In this regard, the basic cluster analysis approach was quite successful. Specifically, it carved up the large number of individual Speeding Episodes (SEs) into sub-groups that had characteristics that could be meaningfully interpreted, and that were largely distinct from each other. The primary speeding types included the following: 1) Speeding that Occurs around Speed-Zone Transitions: This includes Speeding Up, Speed Drop, and Small Increase types of speeding observed in both Seattle and Texas. These SEs typically have short durations, a high maximum speed, and they occur on lower-speed roads. In these cases, the roadway environment in the lower posted speed segment may be similar enough to the higher posted speed limit segment that it supports faster driving. 2) Incidental Speeding: This is the most common type of speeding, and it involves lowexceedance, short-duration episodes that more likely represented the upper bound normal speed maintenance behavior, as opposed to a separate speeding behavior. 3) Casual Speeding: This is a common type of speeding. Although it is similar to Incidental speeding, it involved speeds that were high enough that drivers were likely aware that they were speeding. However, the durations were relatively brief, which suggests that drivers might not persist in this type of speeding for long (e.g., it could include passing behavior). 4) Cruising Speeding: The defining characteristic of this type of speeding was the relatively long duration. While the longer duration increased a driver’s exposure to safety risk, this type of speeding was more likely to occur on controlled-access, high-speed roads, which reduced the likelihood of unexpected hazards. Another notable aspect of this type of speeding is that only a subset of drivers in Seattle engaged in this type of speeding. Specifically, the subset of drivers that had the highest prevalence of Cruising speeding (i.e., 8-25% of their trips) was limited to 10 drivers, representing all demographic groups 5) Aggressive Speeding: This type of speeding was characterized by relatively high speed exceedance, moderate duration, and a high level of speed variability. This cluster only occurred in Seattle and it generally encompassed riskier aspects of speeding than the other clusters. Similar to the Cruising speeding type, the subset of drivers that had the highest prevalence of Aggressive speeding (i.e., 15-50% of their trips) was limited to 10 drivers, representing all demographic groups. The types of speeding listed above were also remarkably consistent across drivers and locations, with five of the six clusters identifiable in both Seattle and Texas Cluster Analysis to Identify Driver Types Suggest that these Types were not Defined Exclusively by Driver Demographics The cluster analysis conducted using individual drivers’ speeding profiles was successful in identifying four different Driver Types, including: 1) Deliberate Speeders: Drivers in this group averaged a higher proportion of Casual and Aggressive SEs, but lower levels of Incidental SEs than other groups. Individuals in this group also had substantially more SEs than those in other groups. In general, these drivers tended to engage in the more aggressive and deliberate types of speeding substantially more than other Driver Types. Deliberate speeders also reported engaging in risky driving behaviors more frequently than others, and they had the most favorable attitudes towards speeding. 2) Typical Speeders: The distribution of SEs within this group basically matched the distribution across all drivers. The Typical Driver Type was also comprised of the largest number of drivers, and Casual speeding was relatively more common in this group. Individuals in this Driver Type also occupied a middle range in terms of average speeding profiles and frequency of SEs. 3) Situational Speeders: This type was distinct in that these drivers had a much higher proportion of the Speeding Up type of speeding than other Driver Types, and they engaged in minimal amounts of Aggressive and Cruising speeding. Overall, this group only engaged in a little more speeding than the Unintentional Driver Type, but they did not share their favorable views regarding not speeding. 4) Unintentional Speeders: This group was comprised primarily of drivers that engaged mostly in Incidental speeding and some Casual speeding, but almost none of the other types of speeding. These drivers also had attitudes and beliefs that were the most favorable towards not speeding, and they likely represent non-speeders. Analyses on the demographic composition of the groups listed above indicated that there were significant differences across the groups. However, the groups differed largely in terms of degree, since all of the demographic categories appeared in each Driver-Type cluster. Further examination of responses to personal inventory questions showed clear trends regarding attitudes and beliefs about speeding were observed across the Driver Types. Additional Data Analyses on the Relationships between Situational Factors and Speeding The second objective in this project was to conduct additional data analyses on the relationships between situational factors and speeding. A challenge for meeting this objective was the lack of situational data available to conduct the corresponding analyses. While we were able to address this objective at a high level, there are many questions that remained unanswered. Some of the key findings from the situational analyses are described below. The first finding from the situational analysis was that the general “riskiness” of different types of speeding was corroborated by the involvement of riskier elements in Speeding Episodes. This analysis was done by examining how the types of speeding differed in terms of the best proxies for safety risk available in the data set (e.g., exceeding posted speed by more than 20 mph, nighttime speeding, etc.). In Seattle, the Aggressive and Cruising speeding types were associated with relatively higher prevalence of risky conditions, and longer exposure durations. In contrast, the Incidental and Casual speeding types were the least likely to involve risky aspects. The general pattern was similar in Texas, except that there was no Aggressive speeding type found there, so the Cruising cluster appeared to be the riskiest type of speeding in the Texas sample of drivers. The second finding generated from the situational analysis was that there was at least anecdotal evidence of location-specific characteristics affecting both the occurrence and non-occurrence of speeding. Locations with posted speed changes, but where the roadway characteristics remained the same across zones, were a common example of where speeding occurred more often. Other locations where we commonly observed higher levels of speeding included those in which the roadway was more open and provided better separations from hazards, including divided roadways and sidewalks that were set apart from the curb. Implications for Safety Countermeasures A key practical implication that stems from this research is that there is converging evidence that the Deliberate speeder group represents a Driver Type that is notably distinct from other groups. In particular, their speeding behaviors are different in that they speed much more frequently and they tend to engage in the more aggressive and deliberate types of speeding, substantially more than other Driver Types. Moreover, individuals in the Deliberate Driver Type also report engaging in risky driving behaviors more frequently than others, and they have the most favorable attitudes towards speeding. The distinctiveness of the Deliberate Driver Type leads to an important practical implication, which is the possibility of specifically directing safety campaigns and countermeasures towards this group. Because their behaviors and attitudes are outside the norm, they can be identified both by their on-road behavior and by using personal inventory items, which may be a practical way to identify drivers from this group. This is also the most critical group to focus on because they engage in the most aggressive type of speeding, likely in conjunction with other risky driving behaviors. Therefore, changing their behavior may have disproportionately large benefits in terms of reducing speeding-related crashes. A related implication pertains to the different types of speeding identified. Some of the analyses suggested that the relative riskiness of the types of speeding may differ. In particular, the Aggressive type in Seattle and Cruising type in Texas more frequently have characteristics that may increase crash risk in comparison to Incidental or even the Casual types of speeding. Further research is needed to more thoroughly characterize each type. However, if it becomes possible to identify distinct speeding behavior types that are linked to increased crash risk, then it opens up the possibility of making those types of speeding an enforcement priority, or more efficiently deploying resources to specifically target those behaviors, rather than the types of speeding that are less dangerous. This could be particularly effective if it is possible to determine where and when the most dangerous types of speeding are most likely to occur. (Author/publisher)
Samenvatting